Customers contact organizations such as telecommunications companies in order to accomplish tasks including ordering a service, requesting repair service, or asking questions about their bill. When a customer contacts the organization, the customer's call is connected to a user interface which may be a customer service representative or any of a number of automated call center systems.
Automated systems include speech enabled call center applications that allow customers to interact with an automated service using voice commands or natural language speech, often referred to as a speech directed dialog and natural language dialog, respectively. One of the advantages of speech based dialogs is that they provide a natural, easy to use, user interface. However, speech recognition systems sometimes fail to recognize a users utterance under certain conditions such as when background noise is present or when the user speaks with a strong accent.
In present systems, speech enabled call center interfaces such as natural language dialogs and directed dialogs as well as non-speech interface systems such as touch tone dialogs encounter difficulty responding to errors that occur when a particular dialog state fails to successfully complete a particular interaction with a user. Often in present systems, after an error is experienced while using an automated system, the system will simply reprompt a user until a successful interaction has taken place. There is often no upper limit placed on the number of reprompts a user experiences in a single phone call. This often results in a dialogue in which customers are repeatedly asked to restate an utterance leading to considerable dissatisfaction on the part of the user.
Also, many speech system reprompts allow only one method of input from the user (e.g., only natural language, or only speech directed dialog) or the user can only use speech input (the system does not allow touch tone). This is problematic when speech input is inappropriate due to high background noise, or when a user needs help with the natural language response, or when they prefer not to use speech input.